The Tool You're Allowed to Use
Everything I've written about working with agents — the map , where to start , what engineers get wrong — quietly assumes something that isn't true for a lot of people: that you chose your tool. I'm freelance. I pick what I use, I change it when something better appears, and the cost of being wrong is mine. That's an unusual position, and writing as though it were universal is a blind spot I'd…
The article titled "The Tool You're Allowed to Use" discusses two main situations regarding the use of AI agents: when one has influence over what gets adopted and when they do not. The main focus is on persuading others to change their tool choice or adapting existing processes to work with limited tools.
The primary concern is that companies often cite cost as the reason for not adopting better AI tools, but the actual objection is usually privacy-related. Companies are more comfortable sharing their data with external providers when they know the provider does not retain or train models on that data. However, these assurances can change, and relying on these assurances is risky.
For those in a position to influence tool choice, the article suggests reframing the conversation around data retention policies and procurement terms, emphasizing that sending data to third parties is already a common practice for companies. The focus should be on negotiating acceptable terms rather than arguing for the absence of risk.
When faced with limited tools, the article advises breaking down work into smaller units and utilizing available scaffolding, such as standing instructions and tooling. Verification processes become more important and expensive, and investing in mechanisms to automate checks is crucial. The process of specifying, building, checking, and correcting remains the same regardless of the tool being used.
The core takeaway is that the overall methodology for using AI agents remains the same, but the pace at which work can be completed is affected by the capabilities of the tool. Companies with more capable agents and permission to use them will produce higher-quality results more efficiently than those with limited tools. This disparity in output quality is not a reflection of individual talent but rather a structural issue that needs to be addressed.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
